Telcon: 2016 03 10
Massimiliano Culpo edited this page Mar 11, 2016
·
23 revisions
- Ben Boeckel (Kitware)
- Brandon Cook (NERSC)
- Chris Fuson (OLCF)
- D'jay Deo (Kitware)
- Greg Becker (LLNL)
- Greg Lee (LLNL)
- James Amundsen (Fermilab)
- Jake Wynne (OLCF)
- Jim Galarowicz (Krell)
- Joaquin Correa (NERSC)
- Mario Melara (NERSC)
- Matt Legendre (LLNL)
- Patrick Gartung (Fermilab)
- Ray Loy (ALCF)
- Peter Scheibel (LLNL)
- Robert French (OLCF)
- Todd Gamblin (LLNL)
- William Scullin (ALCF)
- External packages (PR #120) finally merged. (Matt Legendre)
* Needs to get exercised a lot before we cut a release -- have at it!
- Still working on an issue with specifying providers * Docs are here (probably need to make concretization policies more prominent):
- Concretization policies
-
External packages
* Group agreed that the
nobuild: True
preference should be switched tobuild: False
- Cray & new platforms support merge (Greg Becker)
- Issues with Intel compilers fixed in #451 (Massimilliano Culpo, Adam Stewart, Luigi Calori)
- Fixed a number of quoting/version check issues with
icc
in Spack's compiler wrapper
- Fixed a number of quoting/version check issues with
- Build dependencies #378 (Ben Boeckel)
- Once this is in, we can think about adding mechanisms to detect external build dependencies.
- potentially have OS and/or RPMs install external package descriptions
- RPM status (Peter Scheibel)
* Peter mostly working on coding up support for
DESTDIR
in many different packages.- Trying out
auto-destdir
; not as painless as he'd like it - Looks like packages that do not support
DESTDIR
are always going to require a bit of extra work to get them staged properly. - Peter is working on a document to describe how we'll project from spack hash space to RPM namespace
- Trying out
-
Spack usage across DOE labs (NERSC, ALCF, OLCF, Fermi, others?)
-
Joaquin & Richard Gerber curious how Spack/packaging tools are used at other sites
-
ORNL, Robert French:
- ORNL currently has infrastructure based on Smithy, looking to move to Spack
- Waiting on Cray modules & support from Mario and Greg (but have a lot of faith :)
-
Fermilab, Jim Amundson:
- Fermilab running large server farms for physics experiments, slightly different model from DOE
- Different people factor systems different ways
- Multiple build tools
- Most tied to "UPS", kind of like environment modules.
- Use UPS to figure out what to build
- CMake, other macros are intimately tied to UPS
- Robert: Similar to ORNL in that people rely on env modules to find deps.
-
ANL, William Scullin:
- ANL has 3 internal compute centers, many systems for deploying libraries.
- LCRC using Spack in production
- Internal advocacy at ALCF
- Last year: joint cluster challenge project with William Scullin and IIT students
- Joaquin: how does Spack funding work?
- LLNL: ~2 FTE total, ProTools team dedicated to long-term support.
- Todd: lead developer/PI
- Matt, others in dev environment group contribute
- Greg Becker: .5 FTE
- Peter Scheibel: .5 FTE
- Ben & DJ at Kitware: .5 FTE
- ORNL: .5 FTE total to spend on Spack
- NERSC: 1 FTE currently (Mario), spinning up another to work part-time (Brandon)
- ANL: Adam Stewart at LCRC has many contributions.
- EPFL: .5 FTE (Massimiliano Culpo) committed to using Spack in production on EPFL systems
- Luigi Calori at CINECA also investigating.
- Other contributors:
- Erik Schnetter at Permieter Institute contributing a lot of Mac support
- Elizabeth Fischer, NASA Goddard recently joined
- Joaquin: long-term it would be really good to have some kind of co-funding structure
-
what should the license be?
-
as long as support stays in place NERSC is happy to be on board
-
Matt: LLNL has some ASC funding (ATDM) for the Production Tools (ProTools) team
- Matt leads this team.
- Team is intended to support LLNL code teams long-term
- Team is also scoped to support community, with hope of getting external participation & contributions
- Long-term software lifecycle.
-
Joaquin: what about hackathons, training?
- NERSC has BES hackathons that have worked well
- Robert: ORNL has hackathons for OpenACC, etc.
- model has worked well
- internal training & hackathons are planned
- DOE-wide hackathons also interesting idea.
-
Jim Amundson: As an end-user, having a Spack formula that could work across ALCF, OLCF, and NERSC would be huge
- Reducing porting effort could save ~1 FTE/year
- Greg Becker: Many packages in Spack have required modifications to work on cray
- e.g.: linker option defaults, etc.
- If you want to port to Cray you may need a lot of conditionals right now
- Long-term we'd like to make packages relatively portable
-
Todd:
- We are looking into broader DOE funding, not clear how much we can get that way but we will try.
- Even if that does not come through, it sounds like we could do this organically and have regular meetings somewhere to work on co-developing packages and features.
- Spackathons?
- Facilities are already putting skin in the game, so community building (not necessarily co-funding) is needed.
- number of people on the call (18) is an indicator that we can keep this going.
- Todd will work on getting meetings/hackathons started.
- Fermi and ANL have limited travel budget so maybe Chicago area?
- ORNL will continue to have smaller cray on-site
- Robert: do you know the site
http://whyisgitbetterthanx.com
?- would be nice to have one of these for spack & other build systems.
- Need to justify switch to Spack to ORNL.